MIT neuroscientists announced on Mar. 16 that they have identified how the brain can focus on a single voice among many, addressing the longstanding 'cocktail party problem.' The research, published in Nature Human Behavior, used computational models to show that amplifying neural activity related to features of a target voice allows it to stand out from background noise.
This finding is significant because it helps explain how people are able to follow conversations in crowded environments where multiple voices compete for attention. Understanding this process could lead to improvements in hearing aids and cochlear implants.
Josh McDermott, professor of brain and cognitive sciences at MIT and senior author of the study, said, "That simple motif is enough to cause much of the phenotype of human auditory attention to emerge, and the model ends up reproducing a very wide range of human attentional behaviors for sound." Ian Griffith, lead author and graduate student at Harvard, added, "The responses of neurons tuned to features that are in the target of attention get scaled up. Those effects have been known for a very long time, but what's been unclear is whether that effect is sufficient to explain what happens when you're trying to pay attention to a voice or selectively attend to one object."
The team modified an existing neural network model so its processing units could be boosted or reduced depending on features like pitch. When given an audio cue representing the target voice's characteristics before hearing a mixture of voices, the model amplified relevant neural activations. This allowed it to identify words spoken by the target voice with accuracy similar to humans. The researchers found that both pitch and spatial location helped improve attentional selection.
McDermott said their model also served as an engine for discovery: "You can use the model as a way to screen large numbers of conditions to look for interesting patterns, and then once you find something interesting, you can go and do the experiment in humans." The team observed that separating voices horizontally made it easier for both humans and their model to focus on one speaker compared with vertical separation.
Looking ahead, researchers hope this approach will help simulate listening through cochlear implants and ultimately improve devices so users can better focus their attention in noisy environments.
